#
#asahi-gpu · 2026-01-29 02:39
alyssanice catch.
#asahi-gpu · 2026-01-29 02:39
alyssayes that's broken then.
#asahi-gpu · 2026-01-29 02:39
alyssathat's awkward.
#asahi-gpu · 2026-01-29 02:39
alyssaoh.. I see what you mean.
#asahi-gpu · 2026-01-29 02:38
alyssaNot saying the code is right but it doesn't look wrong to me
#asahi-gpu · 2026-01-29 02:38
alyssaolivial: Curious what you think the bug is exactly...?
#asahi-gpu · 2026-01-29 02:37
alyssaolivial: if there's no CTS for an XFB related feature it's surely broken :p
#asahi-gpu · 2026-01-29 02:37
alyssaoliver: *
#asahi-gpu · 2026-01-29 02:36
alyssaolivial: the fact AGX can't read interleaved means we effectively can't use it for anything in Mesa.
#asahi-gpu · 2026-01-28 23:26
olivial[1]: https://gitlab.freedesktop.org/mesa/mesa/-/blob/2d38da94d4829b339f580e3170744936c3d8a0dd/src/asahi/vulkan/hk_cmd_draw.c#L3225
#asahi-gpu · 2026-01-28 23:26
olivialbecause otherwise, I think there is a bug here[1], where the application binds multiple VS with different XFB configurations consectitively
#asahi-gpu · 2026-01-28 23:25
olivialfor passthrough GS, is there something that causes hk to unbind the passthrough GS when a new application GS is bound?
#asahi-gpu · 2026-01-28 21:35
oliver*directly = correctly
#asahi-gpu · 2026-01-28 21:35
oliverYea. That's fine for me. I am pretty sure that I set the compression info and plane data directly for the rendered image. Compression artifacts are at the "expected places" when I scan out with interchange format. If I scanout with linear I see that only the first few percent of the buffer are filled with color data. I would expect the full buffer to be filled (sparingly).. I have some comparison pictures from Mac traces.
#asahi-gpu · 2026-01-28 21:32
JamesCalligeros[m]agx wont read interchange fwiw only write it
#asahi-gpu · 2026-01-28 21:31
oliverI'll send pictures tomorrow. Couldn't get agx to work either, it just crashes.
#asahi-gpu · 2026-01-28 21:30
JamesCalligeros[m]DCP just wasnt decompressing the texture
#asahi-gpu · 2026-01-28 21:30
JamesCalligeros[m]A picture might be useful. I saw similar when the compression metadata settings were wrong
#asahi-gpu · 2026-01-28 21:29
jannauchadmed was claiming that dcp supports the agx modifiers as well but had trouble filling the compression metadata
#asahi-gpu · 2026-01-28 21:28
jannaudoes the amount of data change if you render a more complex (less compressible) scene?
#asahi-gpu · 2026-01-28 21:25
oliveralyssa: I am trying to render into a texture with layout agx interchange and scan that out with the dcp. Thing is, everything works mostly. I can see the shape of my triangle due to decompression errors, so I know that the gpu writes the acceleration buffer correctly. The only thing that is totally broken is that color data is only being written into roughly the first few percent of the image buffer. Any idea why that might be?
#asahi-gpu · 2026-01-17 17:26
karolherbstmight be good to verify this won't break anything on LLVM-21 or older and llvm-22 and newer: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39374
#asahi-gpu · 2026-01-13 21:28
alyssaP is for the A-series SoCs
#asahi-gpu · 2026-01-13 21:18
kernel_devThank you! Very much appreciated - I'll look into that P variant some more :D
#asahi-gpu · 2026-01-13 21:11
chaos_princessphone(???), base, pro, max, ultra
#asahi-gpu · 2026-01-13 21:09
kernel_devHi, hopefully it's okay to ask this here: but I was wondering what GPU Variant actually means when talking about Apple's ARM64 graphics units? I see values such as P, G, S, C and D - but am not certain what they actually mean.
#asahi-gpu · 2026-01-07 20:59
jannaui.e. at least one of the options was enabled
#asahi-gpu · 2026-01-07 20:58
jannaubigmac: eric tested gles1,no-gles2,no-opengl / no-gles1,gles2,no-opengl / gles1,gles2,opengl / gles1,gles2,no-opengl / no-gles1,no-gles2,opengl
#asahi-gpu · 2026-01-07 20:54
bigmacOkay, thank you regardless. I see eric used opengl=false in the tests, so I went with it. https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21343#note_1774888. jannau: welp, maybe that's what I'll figure out
#asahi-gpu · 2026-01-07 20:53
jannaualso I don't really expect much (if any) differences from replacing the direct comparissons with the helper calls
#asahi-gpu · 2026-01-07 20:51
alyssa(and people there may be able to help you better than me and jannau)
#asahi-gpu · 2026-01-07 20:50
alyssaAlso, this conversation is probably better moved to #dri-devel since there's nothing asahi specific here
#asahi-gpu · 2026-01-07 20:50
alyssaI don't understand why we would disable opengl when trying to build opengl
#asahi-gpu · 2026-01-07 20:49
bigmacI am trying to measure the before and after sizes as ajax requested here https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38981#note_3259709. Eric did the tests with disabling opengl, but I did not realize that doing that here will not generate build/src/egl/libEGL_mesa.so at all
#asahi-gpu · 2026-01-07 20:49
jannaualyssa: fixing https://gitlab.freedesktop.org/mesa/mesa/-/issues/8340 it seems
#asahi-gpu · 2026-01-07 20:44
alyssabigmac: what are you trying to do exactly?
#asahi-gpu · 2026-01-07 20:44
bigmacjannau: Oh I see the betrayal, when opengl is set to false it looks for it outside of the devenv
#asahi-gpu · 2026-01-07 20:39
alyssabigmac: glxinfo please
#asahi-gpu · 2026-01-07 20:23
jannaudoes that change if only disable 1 or 2 of those options?
#asahi-gpu · 2026-01-07 20:15
bigmachttps://paste.centos.org/view/f67eaad5 seems to be ignoring the devenv
#asahi-gpu · 2026-01-07 20:08
jannaucan you check with `LD_DEBUG=files glxgears ...` if the files from your devenv and nothing else gets loaded? I don't see how " -D opengl=false -D gles1=disabled -D gles2=disabled" can end up with any working OpenGL *
#asahi-gpu · 2026-01-07 20:02
bigmacjannau: Oh I know, I think we misunderstood each other, I wouldn't think this would have any performance impact. I am curious about sizes too
#asahi-gpu · 2026-01-07 20:01
jannaubigmac: Eric was comparing file sizes of the library and did not disable every OpenGL api
#asahi-gpu · 2026-01-07 19:57
bigmacjannau: Hm, well then I am doing something wrong as it continues to work. meson devenv -C build glxgears -swapinterval 0.5 yields the same FPS as through running it with what's installed
#asahi-gpu · 2026-01-07 19:51
bigmacnoopwafel: Eric measured it at 2.6% 3 years ago in https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21343. 38981 has less changes, but I'd be curious to see
#asahi-gpu · 2026-01-07 19:48
jannauhk is the vulkan driver. the opengl driver is called asahi. I'm not sure what keeps working with (desktop) OpenGL and OpenGL ES 1.x, 2.x and 3.x disabled. I would expect supertuxkart not to work at all
#asahi-gpu · 2026-01-07 19:31
noopwafelI have no clue (although do you really expect any impact from that change?), but why buildtype debug and not debugoptimized (or release)?
#asahi-gpu · 2026-01-07 18:50
bigmachow to properly test it
#asahi-gpu · 2026-01-07 18:50
bigmacWhen I build mesa with `-Dgallium-drivers=asahi -Dvulkan-drivers=asahi --buildtype debug -D opengl=false -D gles1=disabled -D gles2=disabled` should supertuxkart utilizing the opengl driver (set in settings) still run smoothly? I am trying to see whether https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38981 makes any impact. With hk being a gallium driver I am unsure of
#asahi-gpu · 2026-01-02 22:38
AlyssaRosenzweig[m]i know you are but what am i

plus ›